Ghosts in the Machine
Digital Identity & AI Ethics
Twenty-eight young people from four countries spend ten days in Calabria investigating what automated systems have already decided about them — using nothing but the phones they already carry — and write a public manifesto about what must stay human.
The need
A self written by systems you never agreed to
A young person turning eighteen today inherits a self that is partly written by systems they never agreed to and mostly cannot see. Before they apply for their first credit card, a scoring model has already formed an opinion of them. Before they see a job listing, a filter has decided whether they are shown it. Before they scroll past an advertisement, an inference engine has already guessed their mood, their income bracket and what kind of person they would like to believe they are.
The problem this project addresses sits one layer below "digital skills" as usually taught: it is not whether a young person can use an app, but whether they can recognise when an app — or the automated system behind it — is making a decision about who they are.
of 13–14 year-olds across the EU still lack basic digital skills — the understanding that lets a person ask what a system did with their data, let alone contest it.
ICILS 2023is the Union's own 2030 target for that same figure. The Digital Education Action Plan 2021–2027 names this as a structural, not incidental, gap.
European Commissionthe year the EU's Artificial Intelligence Act entered into force — a Union-wide instrument that no single national programme can meaningfully test against lived reality alone.
AI Act, in force 2024"The net is vast and infinite."Ghost in the Shell (1995)
Young people describe this in strikingly consistent terms even without a shared vocabulary for it: a sense that their phone "knows too much", paired with an almost complete inability to say what specific system knows what, or what could be done about it. Left where it is, that gap calcifies into two equally unhelpful postures.
Uncritical surrender
"The algorithm knows best." The system is treated as neutral, accurate and beyond question — so an unfair outcome is absorbed as a personal failure.
Blanket cynicism
"It's all manipulation anyway." Every automated system is written off equally, which gives up on distinguishing a genuinely useful one from an exploitative one.
Neither leaves a young person able to act. And neither can be resolved inside one country's classroom. The systems young people meet — recommendation engines, credit-scoring models, hiring filters, content moderation — are built on data pooled across borders and deployed near-identically from one national market to the next. A participant inside one national media environment has no way of telling whether what they experience is a local quirk or a shared European condition. That distinction only becomes visible when young people from different countries sit down together and compare, item by item, what their own devices are quietly doing to them.
Try it · the Day 2 method
What does the algorithm know about you?
This is the exercise participants run on Day 2, on their own devices, with their own real examples. Tick anything below that you have done this week, and watch the third column fill in — the column no one ever shows you.
Ordinary things you did
What the system builds from it
0Illustrative demonstration built from publicly documented ad-tech and scoring practice. Nothing is collected, stored or transmitted — this runs entirely inside your own browser. The point is not the specific example: it is that a participant who could not previously distinguish "data I gave" from "data inferred about me" can do so unprompted by Day 3.
What the project sets out to do
Five objectives, each with a mechanism attached
Every objective below is tied to a specific session, a specific method and a specific piece of evidence — not asserted in general terms. Open any of them.
The method
Three columns a participant can fill in for themselves
The identity-mapping framework introduced on Day 2 is deliberately small enough to be remembered without notes, and portable enough to survive the end of the exchange. It requires nothing beyond a device the participant already carries.
01
Data given
What did I actually hand over? A tap, a pause, a postcode, a timestamp, a permission granted once and forgotten. Participants list it from their own device rather than from a slide.
02
Data inferred
What was assumed beyond that? A mood, an income band, a life event, a category you were placed in without being told. This is the step almost no one is taught to look for.
03
Decision that followed
What changed as a result? A price, a ranking, an offer withheld, an application that vanished into silence. Naming the decision is what turns unease into a checkable question.
Investigation, not instruction
Nothing in the ten days is delivered as a lecture on what AI is or how algorithms work in the abstract. Every session starts from participants' own devices and their own collected examples.
Their material, their conclusions
Facilitators supply the analytical method and the fairness and transparency criteria. They never supply the conclusions, or the wording participants choose for the Manifesto.
No prior credentials required
No computer-science background, no fluent English, no policy vocabulary. The exercise measures attentiveness to one's own experience, so a participant with no formal exposure contributes as sharply as anyone in the room.
A public artefact, not a private realisation
A private realisation that "the algorithm knows too much" changes very little on its own. The week ends with something a young person did: a signed, public declaration others can read and add their name to.
Programme alignment
Four horizontal priorities, each with its own mechanism
The project operationalises the Action's own aim to help young people "develop skills and attitudes" and "raise awareness about socially relevant topics" by treating AI and data literacy as a practised, testable skill rather than a topic delivered to participants. It addresses all four horizontal dimensions — each through a specific mechanism, not a general statement of allegiance.
Digital transformation
- The priority the project is built around. The method matches the Programme Guide's own framing: participants develop critical understanding of "the risks and opportunities of digital technology" using their own phones as source material — squarely within SALTO's named project focus, "Artificial Intelligence in education, training and youth work".
Inclusion and diversity
- The analytical method requires no specialist vocabulary and no equipment beyond a device participants already use daily. The small number of participants without a personal smartphone are not excluded from the exercise: a shared device is provided through inclusion support, so the method is not built on an assumption of ownership.
Environment and climate
- The physical infrastructure behind "the cloud" — data centres, their energy and water use — is treated as part of the subject matter rather than leaving digital content immaterial in participants' minds. An automated system that infers a person's identity is also, at the same time, a physical industrial process with a real resource cost.
Participation in democratic life
- The central output is a piece of civic authorship addressed to a genuine external audience — any young person who visits the site and signs — rather than an internal reflection exercise. It engages directly with human dignity and privacy as protected values under EU law, including the non-discrimination principle at stake when an automated system produces a biased outcome.
Goal #4's first target calls on young people to become "critical and responsible users and producers of information", and its fifth target calls for "easy access to understandable youth-friendly information that follows codes of ethics and quality standards". An algorithmically generated or curated claim is, functionally, the same failure of verifiable information that Goal 4 was written to address; this project extends that logic from news content to the automated systems shaping what a young person sees, is offered, or is denied.
This is also a gap the European Youth Dialogue has already flagged and not yet filled. The Youth Goals Task Force's review of the eleventh cycle found that while the current eleven Goals "remain relevant", young people themselves identified missing themes for the goals that will replace them — with digitalisation named explicitly (European Commission, Revising the European Youth Goals, December 2025). This project treats that gap as live and current, rather than waiting for it to be closed by a future policy cycle.
Youth Quality Standards area five — sharing knowledge about the Programme itself — is treated as a planned Day 1 session rather than an assumed side effect, so participants leave able to name not only the ethical questions they investigated but the specific EU mechanism, this Action, that funded the investigation.
The programme
Ten days that move in one continuous arc
Not a set of disconnected sessions. The week opens outward — participants map their own data, then compare across four countries — turns to where automated decisions already shape adult life, then turns inward to what should remain theirs, and closes with a public text. Select any day.
Try it · the Day 5 tool
The Ghosts checklist
On Day 5 each team builds and applies a plain checklist for deciding whether a claim made about or by an AI system actually holds up. Four questions, no technical vocabulary, usable on nothing but the device you already carry. Pick a claim and work through it.
"If a computer can generate its own ghost, and create a soul all by itself, what is the point of being human?"The Puppet Master · Ghost in the Shell (1995)
Where & with whom
Four countries meeting in Calabria
The consortium is a complementarity argument, not a list of organisations that happened to be available. Select a node to see what each partner brings.
Venue
Lamezia Terme, Calabria
Dates
01–10 May 2027
Format
Youth Exchange · KA152-YOU
On site
34 people, 8 core activity days
All four organisations are, in Erasmus+ terms, relatively new to this specific Action. The consortium treats that newcomer profile openly as a fact rather than as a gap to conceal: more experienced Group Leaders and Facilitators are recruited across the consortium to provide the pastoral and safeguarding depth the host groups' youth staff are still building. Each partner also inherits a fully worked session design, the identity-mapping framework, the AI-claim checklist and the manifesto-drafting arc — at no development cost of its own.
A training venue in Lamezia Terme with accessible indoor working space for small-group, device-based investigation alongside a plenary space for the Day 9 Manifesto session. Exact address to be confirmed. Land-based green travel is booked wherever journey time and cost make it realistic.
Who is in the room
34 people, four national teams, one shared starting point
The age range is deliberately narrow. 18–30 is the range in which the automated decisions this project examines — credit approval, job screening, housing applications, university admission — start to have direct, concrete consequences in a participant's own life, rather than being an abstraction encountered secondhand.
Recruited outside the usual pools
Selection deliberately bypasses university networks and prior Erasmus+ applicant pools, running instead through local youth centres and community and vocational channels — the places that reach young people the existing networks miss.
No prerequisites, stated plainly
The public call states that no prior Erasmus+ experience, English fluency or technical background is required — only a willingness to look closely at a device you already use every day.
Device ownership is never a filter
Where a selected participant has no personal smartphone, a shared device is arranged through inclusion support and identified at the same point as dietary and accessibility needs — well before arrival, not discovered on Day 2.
Learning outcomes
Four competences, deliberately not eight
A smaller set of genuinely demonstrated competences is stronger than a full set asserted without a matching mechanism. Each of the four below is tied to a named session, a named method, an observable behaviour and a piece of evidence that is actually retained. Youthpass is introduced on Day 1 and used every evening.
Impact
Three levels, each one counted rather than assumed
The dissemination footprint is deliberately limited to what a small consortium can actually sustain: no new venue is booked, no outside guest is invited, and no partner is asked to do anything beyond what it already runs. What is claimed is narrow — and that narrowness is what makes it checkable.
Local
A live document, not a memory of a week abroad
Each participant returns home able to point their own social circle, family and local youth network directly at a live, signable declaration. National teams share it through channels they already run, rather than organising a new standalone public event.
Measured by: visits and signatures attributable to each team's own sharing, logged in the one-month follow-up survey.
Regional & national
A ready-made session format, at no marginal cost
Each organisation folds the identity-mapping method and the Ghosts checklist into its own existing local youth work as a reusable session format. Euroactive Serbia additionally applies its Google for Nonprofits account to run in-kind search advertising pointing regional visitors to the Manifesto.
Measured by: one organisation doing one well-defined thing, rather than every partner taking on outreach work it has not been resourced for.
European
Comparative evidence that did not exist before
A live, participant-assembled record of how algorithmic identity-construction is experienced across four national and platform contexts, plus the Manifesto and the checklist as open, plain-language tools any youth worker can pick up. Uploaded to the Erasmus+ Project Results Platform and SALTO.
Measured by: signatures the Manifesto carries, and youth workers who can be shown through the platform uploads to have accessed the method.
How success is evaluated
Four stages, each mapped back to a numbered objective rather than treated as a single generic feedback exercise.
Daily
Reflection circles
Same-day visibility into whether a session landed as intended, in participants' own qualitative response.
Day 5 · midpoint
Checkpoint
Held straight after the checklist sessions, so facilitators can adjust the Day 6–7 reflective and drafting arc if gaps in understanding surfaced.
Day 9 · on site
Assessment & survey
Can a participant now name two ways a system builds an inference, and describe one criterion for checking an AI claim? Completed in the plenary, so a full response from all 28 is realistic.
+1 month
Follow-up
How many visits and signatures did each team's sharing generate — and has the participant actually applied the checklist to anything since returning home?
The output
The Ghosts in the Machine Manifesto
A public declaration, written collectively in participants' own words, stating what young people believe must remain human, private and freely chosen in an automated environment — published on an open website other young people can read and add their name to.
The website is coded and structurally built by the project team before anyone travels — page structure, signing mechanism, basic styling all in place. That is deliberate sequencing: it means the on-site Day 8 session is spent on content and voice, not on solving technical problems under time pressure.
The clauses stay empty here on purpose. Writing them in advance, on the participants' behalf, would contradict the one thing the project is built to test: whether a youth exchange can treat participants as co-authors of a public document from day one, rather than as an audience for content a facilitator has already prepared.
Survives the grant
Stays open for signature indefinitely at no ongoing cost, needing no further funding, staff time or specialist maintenance to stay live.
Independent of the consortium
Open-licensed and hosted independently of any single organisation, so a youth worker in a fifth country can use the method two years from now without contacting anyone who built it.
Authored, not summarised
All 28 participants contribute at least one authored element across Days 6–8 — a clause, a short recorded statement, or a piece of written reflection. No staff-produced summary stands in for it.
How it is run
Safety, management and sustainability, with checks attached
Quality Criterion 3 asks whether safety measures are adequate and effective, not only whether they are foreseen — so each vector below carries a check rather than standing as an unverified promise.
Safety & protection
- Physical. Day 1 venue walkthrough and emergency briefing before any working session. Group Leaders confirm by show of hands that every participant knows the nearest exit, the first-aid kit and the named on-site contact.
- Psychological. An explicit right to step back from any discussion of personal data or digital habits, without explaining why. Checked directly in the Day 9 anonymous survey: did a participant who wanted to step back feel able to?
- Logistical. A contingency fund for travel disruption, and one clearly identified on-site team member reachable by phone at any hour, named in the pre-departure Infopack.
Management
- Five Quality Standards headings structure responsibility, so each area maps to a named standard rather than an assumed division of labour.
- Exactly two coordinator video calls — ten weeks and two weeks before departure — matching what a two-person coordination team can realistically sustain. Anything urgent between them goes by email, on the record.
- Concrete escalation thresholds. Recruitment more than two weeks behind, gender balance drifting beyond a defined margin, or a budget line deviating by more than ten percent triggers immediate escalation rather than silent absorption.
Environment
- Land-based green travel booked ahead of flights wherever journey time and cost make it realistic, and a fully paperless preparation and delivery workflow.
- The cloud has a physical footprint. Day 3 includes a short factual segment on the energy and water used by the data centres behind automated recommendation — not bolted on, but part of the subject matter.
- Locally sourced catering, chosen deliberately over a lower-cost, higher-footprint alternative. Any physical props are reusable or drawn from what the venue already has.
After it ends
- Organisations integrate the identity-mapping and checklist session designs into their own regular youth programming at no marginal cost — the method relies on participant-owned devices and free tools only.
- Participants keep a hardware-independent habit: it requires nothing beyond the device and the plain criteria they already carry. Checked in the one-month follow-up survey.
- The artefact stays live and open-licensed. Method and output are downloadable from the same Project Results Platform entry, so both audiences are served by one upload.